FB

Fabio Berberi

Engineering Management · Data Science & Optimization

Siena, Italy · EU Citizen

I combine and adapt different algorithms to improve the overall performance of machine learning models.

I continuously experiment with multiple approaches, modifying and combining different methods to observe how changes affect the final results and overall model effectiveness.

8Papers
50+GitHub Projects

About me

I like to bridge mathematical modelling and real-world systems. My work usually starts from a concrete problem (monitoring, planning, anomaly detection), continues with a clear model (optimization, filtering, scheduling) and ends with clean, reproducible code.

I enjoy research-oriented environments, writing papers, and building tools that are both theoretically sound and practically useful. I am particularly interested in Data Science and optimization for complex systems, and I plan to continue along this path in my Master and potentially a PhD.

Skills

Mathematics & Theory

  • Linear algebra (matrices, eigendecomposition)
  • Advanced calculus and mathematical analysis
  • Optimization theory and numerical methods
  • Probability and statistics for ML and time series

Programming & Software

  • Python (NumPy, pandas, scikit-learn, Matplotlib)
  • Jupyter workflows for experiments and analysis
  • JavaScript / TypeScript, Node.js
  • Web development: HTML, CSS, React, Next.js
  • REST APIs and tools with Flask / FastAPI
  • Git & GitHub, clean and reproducible projects

Machine Learning & Data Science

  • Supervised learning (classification and regression)
  • Feature engineering and feature weighting
  • Robust and multi-objective loss design
  • Model evaluation: cross-validation, RMSE, MAE, MAPE
  • Handling missing data and time-series imputation (Kalman, KNN, interpolation)

Optimization & Metaheuristics

  • Particle Swarm Optimization (standard and variants)
  • Grey Wolf Optimizer and Simulated Annealing hybrids
  • Differential Evolution and Genetic Algorithms
  • CMA-ES, Firefly and Bat algorithms
  • Gradient-based methods: Gradient Descent, Adam
  • Nelder–Mead and other derivative-free local search
  • Hybrid and ensemble metaheuristics

Kalman Filtering, Time Series & Feature Weighting

  • Linear, Extended and Unscented Kalman Filters
  • Innovation-based adaptive noise covariance tuning
  • Time-series reconstruction and missing-data imputation
  • Adaptive feature weighting with Kalman-based methods
  • Entropy-guided perturbations to avoid stagnation

Dynamical Systems, Chaos & Control

  • Nonlinear dynamical systems and chaos analysis
  • Chaos metrics: permutation entropy, Lyapunov exponents
  • Optimization of chaotic vs stable regimes with PSO
  • Feedback control and parameter tuning via optimization
  • SIR epidemic models and vaccination control strategies

Operations Research & Scheduling

  • Single-machine and multi-machine scheduling
  • Resource-constrained project scheduling (RCPSP)
  • Project planning and Gantt-based resource allocation
  • Cost-based planning and scheduling heuristics

Research focus

Hybrid Optimization & Intelligent Search

Design of hybrid optimization frameworks that combine swarm intelligence, evolutionary strategies, chaos-inspired mechanisms and Kalman-based ideas to explore complex, high-dimensional search spaces. Focus on adaptive exploration–exploitation balance, domain-based learning and performance-driven fusion of multiple optimizers.

Kalman-Based Learning & Adaptive Model Control

Development of Kalman-driven learning architectures for feature weighting, uncertainty handling and missing-data reconstruction. This includes adaptive noise modeling, innovation-guided optimization and integration of filtering techniques into machine learning pipelines for more robust and explainable models.

Complex Systems, Chaos & Dynamical Optimization

Analysis and control of nonlinear dynamical systems using chaos-aware optimization. Work includes entropy-based chaos metrics, Lyapunov-driven adaptation and parameter tuning for both chaotic and stable regimes in simulated and physical setups, with a focus on controllability and robustness.

Applied Optimization in Safety & Healthcare Systems

Application of optimization and filtering methods to safety-critical and healthcare-related problems: worst-case scenario search in vehicle safety, epidemic vaccination strategies based on SIR models, and monitoring frameworks that reduce communication and computation while preserving reliability and decision quality.

A dedicated section with individual projects will be added soon.

Selected papers

Domain-as-Particle PSO paper
Published

Neural Feature Weighting

Domain-as-Particle with PSO Methods for Neural-Network Feature Weighting

Fabio Berberi, Paolo Mercorelli

A domain-as-particle PSO framework that partitions the feature-weight space into subdomains, combines surrogate modeling with K-Fold validation, and efficiently searches for optimal neural feature weightings on medical data.

  • Multi-domain PSO with domain movement via characteristic vectors
  • Surrogate model to reduce expensive NN evaluations
  • Composite metric combining accuracy, precision and recall
Download PDFOptimizationNeural NetworksMedical Data
Healthcare monitoring paper
Accepted

Healthcare Monitoring

Event-Triggered Robust Kalman Filtering with Firefly Optimization, Conformal Triggering and Sparse Sensor Gating

Fabio Berberi, Paolo Mercorelli

A hybrid Kalman framework for healthcare monitoring that combines robust innovation shaping, Firefly-based parameter optimization, conformal event-triggering and sparse sensor gating to reduce communication and computation while preserving accuracy.

  • Event-triggered updates with statistical guarantees
  • Firefly optimization for robust filter parameters
  • Sparse sensing and sensor gating for low-energy devices
Download PDFKalman FilteringHealthcareMetaheuristics
Crash system paper
Accepted

Vehicle Safety

PSO-Based Optimization for Identifying Worst-Case Crash Scenarios

Fabio Berberi, Paolo Mercorelli

A domain-as-particle PSO model for discovering critical crash scenarios in vehicle systems, exploring the input space through hierarchical swarm domains to find worst-case impacts and unsafe configurations.

  • Search over crash parameters via PSO domains
  • Identification of worst-case vehicle behaviour
  • Framework applicable to safety validation and testing
Download PDFSafetyPSOAutomotive
Vaccination PSO paper
Accepted

Epidemic Control

Optimal Vaccination Strategies for Pandemic Control: A Cost-Driven SIR Model with Domain-as-Particle PSO Optimization

Fabio Berberi, Paolo Mercorelli

A cost-driven SIR epidemic model optimized with a Domain-as-Particle PSO framework, designing vaccination strategies that minimize both intervention time and overall vaccination effort.

  • SIR-based feedback vaccination control law
  • Domain-as-Particle PSO over vaccination parameters
  • Trade-off between effort cost and time to herd immunity
Download PDFPandemic ControlSIR ModelsPSO
Coming Soon...
Kalman imputation paper

Time Series Imputation

Kalman Filter for Missing Data Imputation in Time Series

Fabio Berberi, Paolo Mercorelli

A Kalman-based imputation pipeline that reconstructs missing values in time series, preserving temporal structure and outperforming classical imputers such as interpolation and KNN in RMSE and MAE.

  • Kalman imputation for uni/multivariate time series
  • Comparison with mean, KNN and spline interpolation
  • Robust under MCAR and block-missing patterns
Download PDFKalman FilterMissing DataTime Series
Coming Soon...
CKFW-Net++ feature weighting paper

Feature Weighting

CKFW-Net++: Adaptive Chaotic Kalman Feature Weighting Network

Fabio Berberi, Paolo Mercorelli

A UKF-based feature weighting framework with entropy-guided chaotic perturbations, sparse gating and robust multi-objective loss for noisy and high-dimensional datasets.

  • Adaptive UKF with online Q/R estimation
  • Entropy-driven chaos to avoid stagnation
  • Sparse normalized gates for interpretability
Download PDFFeature WeightingKalmanChaos
Coming Soon...
Chaos optimization paper

Chaos Optimization

Maximizing or Minimizing Chaos in Nonlinear Maps: A Kalman Filter–Guided PSO Approach

Fabio Berberi, Paolo Mercorelli

A Domain-as-Particle PSO where a nonlinear Kalman filter acts as a chaos detector, combining innovation variance, permutation entropy and Lyapunov exponents into a composite chaos metric.

  • Composite chaos metric with Kalman innovations
  • Domain-as-Particle PSO over parameter space
  • Applications to chaos maximization and minimization
Download PDFChaosPSOKalman Filter
Coming Soon...
FusionManagerPPv7 meta-optimization paper

Meta-Optimization

FusionManagerPPv7: Dynamic PSO-Based Weight-Space Fusion of Ten Optimizers

Fabio Berberi, Paolo Mercorelli

A PSO-driven framework that fuses ten optimization algorithms in a continuous weight space, adapting exploration–exploitation balance over classical benchmark landscapes.

  • Outer PSO in 10D weight space
  • Fusion of swarm, evolutionary and gradient-based methods
  • Robust performance across benchmark functions
Download PDFMetaheuristicsEnsemblePSO
✕Zoomed paper

Education

BSc Engineering Management

University of Siena

Focus on operations research, optimization, scheduling, project management and quantitative decision-making.

Erasmus, Research & Industry

Leuphana University (Germany) & SATER

Nine months of Erasmus in Germany, working on optimization and Kalman-based methods with professors, plus an internship at SATER focused on fault detection for production machines and analysis of industrial measurement data.

Next step: Master in Data Science

Planned

Future goal: pursue a Master's degree in Data Science, with a strong focus on optimization, statistical learning and research-oriented projects in complex systems.

Contact

Open to research collaborations, internships and part-time roles in data science, optimization and intelligent monitoring.

f.berberi@student.unisi.itgithub.com/F-universeLinkedIn